Possible incorrect handling of hard case

[jarad-utcs]

Possible incorrect handling of hard case

[cjones6]

Hi! I fixed the imaginary issue; thanks for noticing it and providing an example. Regarding your other comment, Conn, Gould, and Toint (2000) say (p. 181, section 7.3.1.3) that "All is still fine so long as Δ is smaller than the critical value Δcri = ψ(−λ1)." (Δ is r in my code). I haven't looked at this algorithm in a long time, but I'm not sure that it would always need to enter the code corresponding to Step 3b in their Algorithm 7.3.6 when B is indefinite and gamma1 is 0.

Conn, A.R., Gould, N.I.M., Toint, P.L. (2000) Trust-Region Methods. SIAM, Society for Industrial and Applied Mathematics. https://epubs.siam.org/doi/book/10.1137/1.9780898719857

[jarad-utcs]

You're correct. I've benchmarked the subproblem solver and verified it finds optimal solutions for all cases where B is indefinite. Thanks!

[cjones6]

Great, thanks for letting me know!